Research On Human Action Recognition Algorithm Based On Bone Frame

With the continuous development of human-computer interaction,virtual reality,video intelligent monitoring,medical diagnosis and monitoring and motion analysis,human behavior recognition technology has been developing rapidly and has become a research hotspot in the field of computer vision.Now it has a very broad application prospects and very considerable economic value.Human body skeleton frame data has been used to study the simple human action representation and human behavior recognition.The main work is as follows:1)In the recognition of human behavior,how to construct a precise behavior description feature vector is the key of the study.In this paper,Kinect is used to acquire the three-dimensional coordinates of the 20 main joints of the human body.The human skeleton is modeled by the coordinates of the joints.The vector of the human body is constructed and then the vector angle between the vectors is taken as the vector of human behavior description.2)The human behavior recognition algorithm based on dynamic time warping(time warping dynamic,DTW)and improved BP(Propagation Back)neural network is studied.For the DTW algorithm,because different distance algorithm in the algorithm of behavior recognition will produce different results.So we choose the representative Euclidean distance,Mahalanobis distance and chi-square test to study.For the BP neural network,in order to improve the efficiency of the network and enhance the generalization ability of the network,this paper uses two methods to optimize the training data in the experiment.Firstly,according to the proportion of human standard to remove junk data,and then use the factor analysis method to reduce the dimension of training data.3)Finally,two different methods of human behavior recognition are proposed,and different experiments are designed to validate them.Through a lot of experiments,the advantages and disadvantages and applicability of the three distance algorithms in DTW algorithm are studied.According to the experiment of the improved BP neural network,it is proved that the improved BP neural network method can get good recognition effect.The feasibility and validity of the behavior representation and the two behavior recognition methods are demonstrated.